Research Of Binary Features Based On Android Platform

With the rapidly growing popularity of mobile devices and increasing image datasizes, binary features has become a new hotspot since the traditional feature methodsuse high dimensional features of floating-point type, and cost a lot of storage spaceand computing time. They are able to quickly compute and match with efficientstorage.BRIEF(Binary Robust Independent Elementary Features), ORB (Oriented FASTand Rotated BRIEF), BRISK(Binary Robust Invariant Scalable Keypoints) andFREAK (Fast Retina Keypoint) these four binary features were researched in thispaper while SIFT(Scale Invariant Feature Transform) and SURF(Speeded Up RobustFeatures) as references. Firstly, the general characteristics of binary features wereresearched in feature detecting, description and matching. The characteristics includesampling models and robust. The space and time performance was studied both onmobile device and PC. Also, the robustness of binary features under scale change,rotation, blur, illumination change and perspective change was researched. Thedifferent stages of feature methods and their performance bottleneck were analyzed.The performance and platform influences and limitations were studied based on themobile platform. Finally, binary feature methods were paralleled to improve theperformance.Experiments and analysis were done on the automated evaluation system, whichwas designed and implemented on both PC and Android platforms. Evaluation systemwas extended with new metrics. The test image dataset was expended. The resultsshow that binary features have superior performance in time and space that theefficiencies are improved about10times. Also they have very good robustness overallwhich is comparable with SIFT’s. Parallel can achieve a speedup of1.3or more,which can improve the performance.